The second quarter memory semiconductor performance recorded consecutive deficits due to the sluggish recovery in demand for consumer products such as mobile and PCs and general servers. However, a gradual recovery is expected due to the expansion of orders for high value-added lineups such as HBM3 and DDR5 due to the boom in the AI server market.

SK Hynix disclosed its second quarter performance this year during a conference call on the 26th. The company announced that it recorded the following business performance in the second quarter: △Sales of KRW 7.3059 trillion, △Operating loss of KRW 2.8821 trillion, and △Net loss of KRW 2.9879 trillion.

Sales increased 44% compared to the previous quarter, and operating profit showed a 15% decrease in the second quarter compared to the loss of KRW 3.4023 trillion in the first quarter.

On the other hand, net income recorded a net loss of close to 3 trillion won in the second quarter, a 16% decrease from 2.5855 trillion won in the first quarter.

According to SK Hynix, the prices of general DRAMs such as DDR4 continued to fall due to the weakness in the PC and smartphone markets, and general server products also experienced a decrease in demand, which affected performance. However, sales of high-priced, high-spec products used in AI servers increased, leading to an increase in the overall DRAM ASP compared to the first quarter. Accordingly, sales of DRAM and NAND increased in the second quarter, and in particular, the increase in the average selling price (ASP) of DRAM compared to the first quarter had a significant impact on the increase in sales, it explained.

SK Hynix explained, “As the generative AI market centered on ChatGPT has expanded, demand for memory for AI servers has skyrocketed.” “Accordingly, sales of premium products such as HBM3 and DDR5 have increased, and sales in the second quarter increased by 44% compared to the first quarter, while operating loss decreased by 15%.” It was revealed that the graphics process segment, including HBM3, accounts for about 20% of total shipments.

He also mentioned that the operating loss was reduced as inventory valuation losses decreased while continuing company-wide cost-cutting efforts. SK Hynix announced that it recorded an inventory valuation loss of KRW 500 billion in the second quarter, a decrease of nearly half compared to the previous quarter when it was close to KRW 1 trillion.

In the second half of the year, it is expected that inventory reduction will accelerate and prices will stabilize, and the possibility of additional inventory valuation losses is reported to be low. For some products, there may also be foreign exchange gains, which is expected to have a positive effect on profit and loss recovery.

At the performance announcement that day, the company diagnosed that the strong demand for AI memory will continue in the second half of this year and that the effect of memory companies' production cuts will become evident.

Accordingly, SK Hynix plans to steadily increase sales centered on HBM3, AI memory, DDR5 and LPDDR5, high-performance DRAMs, and 176-layer NAND-based SSDs to accelerate the improvement of its performance in the second half.

In addition, the company announced that it will rapidly increase the mass production ratio in the upcoming upturn by improving the initial mass production yield and quality of 10nm-class 5th generation (1b) DRAM and 238-layer NAND this year. However, the company reported that the rate of decrease in NAND inventory was slower than that of DRAM, and decided to expand the scale of reduction in NAND products.

SK Hynix said in a conference call that “NAND has higher industry inventory levels and lower profitability compared to DRAM, so we will reduce production by an additional 5-10%,” and thus expected inventory normalization to come a little sooner.

During the conference call that day, questions about HBM were asked one after another, showing the current boom in the AI server market and the high level of attention paid to HBM products.

In particular, regarding analysts’ concerns about the ‘HBM downtime’ next year, SK Hynix said, “It is difficult to completely deny these concerns if CSPs, platforms, and big tech companies meet demand in the short term,” and added, “However, if we only look at next year, we have secured long-term supply and demand visibility, and in particular, AI servers at this point are focused on training, and we believe that supplementation will be needed for inference servers in the future.”

Accordingly, it is expected that there will be a continuous demand for inference server expansion from companies in demand for AI servers, and in terms of demand-based aspects, AI services face the task of evolving into customized services, so if connected to the commerce sector, this could act as an underlying demand and affect the continued demand for HBM.